Automated Image Analysis Software for Quality Assurance of a Radiotherapy CT Simulator

Automated Image Analysis Software for Quality Assurance of a

Radiotherapy CT Simulator

Andrew J Reilly

Imaging PhysicistOncology PhysicsEdinburgh Cancer CentreWestern General HospitalEDINBURGH EH4 2XU

Phone: 0131 537 1161Fax: 0131 537 1092E-Mail: andrew.reilly@luht.scot.nhs.ukWeb: http://www.oncphys.ed.ac.uk

Overview

• Radiotherapy imaging

• RT Imaging QA: problems and solution

• Describe features of auto analysis software

• Demonstrate application to CT-Sim and Sim-CT

• Outline experience to date

Imaging Modalities for RT

• Common• Simulator (fluoroscopy)

• CT-simulator

• Digitally Reconstructed Radiographs (DRRs)

• Simulator-CT (single slice and cone-beam)

• Electronic Portal Imaging Devices (EPIDs)

• ‘Emerging’• Ultrasound

• MRI

• PET

• On treatment cone-beam CT and kV radiography

Integrated System

RT Imaging QA: Essential Tests

• Geometric Accuracy in 3D• In and out of image plane (pixel size, couch travel)• Mechanical alignments• Laser alignment

• Image quality• Sufficient for purpose?• Consistent over time

• Accurate physical information• CT number / HU calibration -> electron density

• Testing of overall system• Geometrical co-registration• Transfer of image data

The Problems…

• Different tests are specified for different modalities

• Range of ‘equivalent’ test objects

• Most tests are only semi-quantitative

• Operator dependency

• Frequent (daily/fortnightly) comprehensive testing is required BUT most tests are time-consuming

• Some imaging equipment performs too well!

• Difficult to test integrated system.

The Solution…

• Develop single, uniform approach for all RT imaging modalities• + display devices, film processors, etc.

• Robust, fully objective and quantitative

• Analysis performed by computer

• Results automatically stored in database for trend analysis, etc.

The Approach

1. Develop Appropriate Phantom

Signal s1

s2

SNRin = s1 / s2

2. Acquire Image of Phantom

Signal s1s2

SNRout = s1 / s2

fDQESNR

SNR

in

out

2

Determining the DQE

NPS

MTFDKfDQE

2

Modulation Transfer Function (Phantom)

Noise Power Spectrum (Phantom)

Dose and acquisitionsetting dependent.

Varian Ximatron EX Sim-CT

AdditionalCollimators

Varian Performance Phantom

WATER

MTF

LUNGINNERBONE

CORTBONE

AIR

A

P

R L

A

P

R L

1 2

3

MTF

INNERBONE WATER

A

P

L R

P

R L

A

1 2

3

Varian Uniformity Phantoms

44 cm 34 cm

Polyurethane CastingHU -580

Geometry: Phantom Alignment

• Detect phantom edge• Threshold at –580

• Trace edges and choose largest contour

• Calculate COM

• Compare against CT zero position

Geometry: Pixel Size

• Measure distance between holes

• Use centre of phantom and expected pixel size to identify ‘seek area’

• Local minimum is centre of hole

A

P

R L

1 2

3

A

P

R L

1 2

3

Hounsfield Unit Calibration

-1500

-1000

-500

0

500

1000

1500

2000

2500

0.0 0.5 1.0 1.5 2.0 2.5

Electron Density Rel to Water

CT

Nu

mb

er

ICRU 42

Ax, 80kV, 150mA

Ax, 80kV, 300mA

Ax, 120kV, 150mA

Ax, 120kV, 300mA

Ax, 140kV, 150mA

Ax, 140kV, 250mA

Baseline Values Measured During Commissioning

Hounsfield Unit Calibration

W ATER

MTF

LUNGSOFTBONE

HARDBONE

AIR

A

P

R L

• Calculate from impulse object

Modulation Transfer Function

xPSFFTfMTF

Finite size(DSF) xDSFxPSFxOSF

xDSFFTxPSFFTxOSFFT

xDSFFT

xOSFFTfMTF

Calculation from Impulse Object

Object Spread Function(From ALL pixels in ROI)

Uniformity Phantom Analysis

• Define Useful FOV (UFOV) as 90% FOV

• Calculate:

mean

dev std Variation, oft Coefficien CoV

mean

meanpU

max y, UniformitIntegral

mean

meanpU

min y, UniformitIntegral

mean

pU d

max y, UniformitalDifferenti

1000 Index, Uniformity

peripherycentreUCT

Uniformity Phantom Analysis

Uniformity ProfilesCT Sim: 50 cm FOV

Sim-CT

Urethane Norm

Air Norm

Noise Power Spectrum

• Region of Interest from Uniformity Phantom

• Remove DC component (subtract mean value)

• Perform 2D FFT

• Separation of stochastic noise

area

vuvuvuNPS

22 ,Im,Re,

n

ROINPS

n

NPSNPS nn

s

NPS Example• 100 images of Uniformity Phantom, 50 cm FOV

Production of DRRs• Ray trace from virtual source of x-rays through stack of

CT slices and model attenuation of beam.

SAD100 cm

isocentre

Imaging Plane

X-ray source

Reference: Milickovic et al,Physics in Medicine and Biology (2000) 45:10;2787-2800

Projected backto isocentre

DRR Production Example

CT Slices3D array of voxels

DRR

Edinburgh DRR Phantom

Software Demo

Experience & Conclusions

• New approach appears complicated, but…• Significantly faster than previous methods• More robust, fully objective and quantitative• Greater confidence in results• New ability to follow trends

• Need to finalise DRR phantom• Expand to include other RT imaging modalities